Early warning method for large-scale new type electrochemical energy storage fault

Abstract

This invention provides a novel early warning method for large-scale electrochemical energy storage faults, relating to the field of electrochemical energy storage technology. The steps of this invention are as follows: S1, collecting multi-dimensional correlation parameters and performing preprocessing operations; S2, screening soft-sensor estimation parameters for early fault warning, and using the screened soft-sensor estimation parameters as relevant early warning parameters for fault types; S3, performing early fault judgment; S4, performing early fault location and classification operations; S5, performing early fault warning. The beneficial effects of this invention are as follows: increasing the collection frequency of weak fault-sensitive parameters and reducing the collection frequency of conventional parameters can reduce energy loss; screening soft-sensor parameters based on linear correlation values, non-linear correlation values, and fault sensitivity values ​​improves the accuracy and comprehensiveness of the screening; setting the early warning threshold as a dynamic threshold can adapt to the sensitivity of different relevant early warning parameters to faults; the classification and location method can achieve efficient fault location.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage technology, specifically to a method for early warning of faults in large-scale novel electrochemical energy storage systems. Background Technology

[0002] Large-scale electrochemical energy storage battery systems in new energy bases typically have tens of millions of individual cells. As a result, there are numerous root causes and complex coupling factors that can lead to safety accidents. Relying solely on real-time sensor data has a low correlation with potential faults under actual operating conditions. Therefore, safety early warning algorithms for electrochemical energy storage battery systems built on this basis are difficult to achieve accurate and timely warnings, which can easily lead to the continuous accumulation of risks. Thus, it is particularly important to provide early warnings and address faults in their early stages.

[0003] However, early warning of faults also presents the following challenges: (1) The range of parameter changes is very small. For example, the voltage fluctuation of a single cell is less than 50mV and the internal resistance increases by less than 2%. These minute parameter changes are far below the threshold of serious faults. (2) The changes in parameters may be intermittent and will automatically recover after a period of time; (3) Subtle changes in a single parameter cannot reflect the health status of the battery cell; Meanwhile, factors such as sensor noise, natural aging of batteries, and the large number of individual battery cells can interfere with early warning of faults, further increasing the difficulty of early warning.

[0004] Therefore, there is an urgent need for a method that can solve the above problems and achieve accurate and timely early warning of faults in large-scale new electrochemical energy storage. Summary of the Invention

[0005] The main objective of this invention is to provide a novel early warning method for large-scale electrochemical energy storage faults, thereby solving the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a novel large-scale electrochemical energy storage fault early warning method, comprising the following steps: S1. Collect multidimensional relevant parameters and perform preprocessing operations; S2. Filter the soft sensor estimation parameters for early fault warning and use the filtered soft sensor estimation parameters as relevant early warning parameters for fault types. S3. Perform early fault diagnosis; S4. Perform early fault location and classification operations; S5. Provide early fault warnings.

[0007] Furthermore, the data collection targets three types of energy storage units: lithium-ion battery systems, sodium-ion battery systems, and vanadium redox flow battery systems. The parameters collected by lithium-ion battery systems and sodium-ion battery systems can be categorized into: electrical parameters, thermal parameters, gas parameters, and mechanical parameters. The parameters collected by the vanadium redox flow battery system can be divided into: stack electrical parameters, fluid parameters, ion concentration parameters, and equipment status parameters. The parameters are divided into sensitive parameters for minor faults and normal parameters.

[0008] Furthermore, the parameter sampling frequency adjustment process is as follows: Set the standard value of the sampling frequency, set the sampling frequency of the weak fault sensitive parameter to 1.2 times the standard value, and set the sampling frequency of the normal parameter to 0.8 times the standard value.

[0009] Furthermore, the soft sensing estimation parameters are divided into lithium-ion soft sensing estimation parameters, sodium-ion soft sensing estimation parameters, and all-vanadium liquid soft sensing estimation parameters. Early failures can be categorized as: lithium-ion failures, sodium-ion failures, and all-vanadium liquid flow failures.

[0010] Furthermore, the selection criteria are linear correlation values, nonlinear correlation values, and fault sensitivity values; for any type of fault, the process for selecting the relevant soft-sensor estimation parameters is as follows: The expression for the linear correlation value is as follows: (1); in, The linear correlation value, For covariance, Standard deviation, For the first Time series of soft sensing estimation parameters For the first Time series of severity scores for each fault type, and They correspond in time sequence; The expression for the nonlinear correlation value is as follows: (2); in, This is a non-linear correlation value. The value of the parameters estimated by the soft sensing at a certain moment in the time series. The value at a specific moment in the time series representing the severity score of the fault type. for The ratio of the number of occurrences to the total number of elements in the time series. for The ratio of the number of occurrences to the total number of elements in the time series. For the same time, the soft sensing estimation parameters are taken Severity score The ratio of the number of occurrences to the total number of elements in the time series; The expression for the fault sensitivity value is as follows: (3); in, For the first The first soft sensing estimation parameter is related to the first... Fault sensitivity values ​​for different fault types Estimate the change in parameters for the corresponding soft sensing, i.e. The difference between the maximum and minimum values. This represents the change under normal aging conditions. This represents the change under fluctuating operating conditions.

[0011] Further, the screening process is as follows: Set linear judgment threshold, nonlinear judgment threshold and sensitivity judgment threshold, and use the soft sensing estimation parameters that simultaneously satisfy: linear correlation value greater than linear judgment threshold, nonlinear correlation value greater than nonlinear judgment threshold, and fault sensitivity value greater than sensitivity judgment threshold as the relevant early warning parameters for this fault type.

[0012] Furthermore, the detailed process of step S3 is as follows: S301. For any fault type of any battery, set a warning threshold for each relevant warning parameter; S302. Perform early fault judgment; the condition for early fault judgment is: when at least two relevant warning parameters exceed the warning threshold, it is judged as an early fault.

[0013] Furthermore, the warning threshold is a dynamic threshold, expressed as follows: (4); in, As the warning threshold, For early warning adjustment items, As the baseline value for early warning, , These are the adjustment weights and the base weights, respectively. The expression for the early warning adjustment item is as follows: (5); in, This refers to the fault sensitivity value of the relevant warning parameter for this type of fault. This is the amplitude adjustment coefficient; The baseline value for the early warning is determined based on the values ​​of the relevant early warning parameters during fault-free operation.

[0014] Furthermore, early fault location employs a hierarchical location method, which can be divided into: cluster-level location, module-level location, and cell-level location. Cluster-level location is determined based on cluster-level anomaly scores, the expression for which is as follows: (6); in, For cluster-level anomaly scores, For the first The average value of relevant early warning parameters over several sampling periods; for the function , there is: when When the warning threshold of the relevant warning parameter is not exceeded, Take 0, when When the warning threshold of the relevant warning parameter is exceeded Pick The absolute value of the difference between the warning threshold and the relevant warning parameter; Compare the cluster-level anomaly scores of all clusters, and the cluster with the highest cluster-level anomaly score is the cluster where the fault occurs; After locating the cluster where the fault occurs, module-level localization is performed. Module-level localization is determined based on the module anomaly score, which is expressed as follows: (7); in, For module anomaly scores, , These are the standard deviations of cell voltage and temperature within the module, respectively. This represents the relative rate of change of the module's equivalent internal resistance. , , These are the voltage evaluation weight, temperature evaluation weight, and internal resistance evaluation weight of the module, respectively. Compare the module anomaly scores of all modules, and the cluster with the highest module anomaly score is the module where the fault occurs; After locating the faulty module, cell-level localization is performed. Cell-level localization is based on cell anomaly scores, which are expressed as follows: (8); in, Score for cell abnormalities. , , These are the cell's voltage deviation rate, temperature rise rate, and relative change rate of internal resistance, respectively. , , These are the voltage weight, temperature weight, and internal resistance weight of the battery cell, respectively. Comparing the cell anomaly scores of all cells, the cell with the highest anomaly score is the cell that failed early.

[0015] Furthermore, an early fault warning adopts a tiered warning method, specifically: an anomaly threshold is set. When the cell anomaly score of an early faulty cell exceeds the threshold, a message is pushed to the power plant operation and maintenance manager to carry out operation and maintenance processing on the early faulty cell; when the cell anomaly score of an early faulty cell does not exceed the threshold, a message is pushed to the power plant operation and maintenance manager, and the sampling frequency of the cell data is increased. An observation period is set. If the cell anomaly score of the cell does not improve during the observation period, the normal sampling frequency is restored; otherwise, a message is pushed to the power plant operation and maintenance manager to carry out operation and maintenance processing on the cell.

[0016] Beneficial effects: (1) Increasing the acquisition frequency of sensitive parameters for weak faults and decreasing the acquisition frequency of conventional parameters can reduce energy consumption and costs. (2) Soft sensing parameters are screened based on linear correlation values, nonlinear correlation values ​​and fault sensitivity values ​​to improve the accuracy and comprehensiveness of the screening; (3) Set the warning threshold as a dynamic threshold and introduce a fault sensitivity value for adjustment. This can adapt to the sensitivity of different relevant warning parameters to the fault and avoid the problem of false alarms with fixed thresholds. (4) The hierarchical location method can achieve efficient fault location. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the steps of the method of the present invention. Detailed Implementation

[0018] Example 1 like Figure 1 As shown, a novel early warning method for large-scale electrochemical energy storage faults includes the following steps: S1. Collect multi-dimensional relevant parameters and perform preprocessing operations; the objects of collection are three types of energy storage units, namely: lithium-ion battery system, sodium-ion battery system, and vanadium redox flow battery system. The parameters collected by lithium-ion battery systems and sodium-ion battery systems can be categorized into: electrical parameters, thermal parameters, gas parameters, and mechanical parameters. Electrical parameters include: instantaneous values ​​of individual unit voltage, voltage fluctuation amplitude, instantaneous values ​​of charging and discharging current, current fluctuation coefficient, dynamic value of DC internal resistance, amplitude and phase of AC impedance, cluster voltage deviation, and inter-cabinet voltage difference; Thermal parameters include: cell surface temperature, module temperature gradient, cluster temperature field distribution, ambient temperature, and thermal diffusion rate; Gas parameters include: concentrations of carbon monoxide, hydrogen, carbon dioxide, and volatile organic compounds; Mechanical parameters include: cell casing strain, module vibration frequency and amplitude, structural stress, and pressure changes inside the cabinet; The parameters collected by the vanadium redox flow battery system can be divided into: stack electrical parameters, fluid parameters, ion concentration parameters, and equipment status parameters. The electrical parameters of the fuel cell stack include: single cell voltage, total stack voltage, charge and discharge current, equivalent internal resistance of the stack, positive and negative electrolyte potentials, and polarization voltage; Fluid parameters include: electrolyte flow rate, pipeline pressure, inlet and outlet pressure difference, tank liquid level, electrolyte temperature, electrolyte viscosity, and flow velocity distribution uniformity. Ion concentration parameters include: vanadium ion concentration at the positive electrode, vanadium ion concentration at the negative electrode, ion concentration difference, electrolyte acidity, and ion permeation across the membrane. Equipment status parameters include: pump operating current, pump speed, pump body vibration, leakage signal, valve opening, membrane impedance, and pipeline blockage degree; The above parameters can be divided into sensitive parameters for minor faults and conventional parameters. Sensitive parameters for minor faults include: voltage fluctuation amplitude, voltage fluctuation magnitude, current fluctuation coefficient, dynamic value of DC internal resistance, amplitude and phase of AC impedance, cluster voltage deviation, inter-cabinet voltage difference, module temperature gradient, temperature field distribution within the cluster, thermal diffusion rate, concentrations of carbon monoxide, hydrogen, and volatile organic compounds, cell casing strain, structural stress, pressure change within the cabinet, equivalent internal resistance of the battery stack, electrolyte potential of the positive and negative electrodes, polarization voltage, inlet and outlet pressure difference, electrolyte viscosity, flow rate distribution uniformity, ion concentration difference, ion permeation across the membrane, and degree of blockage in the membrane impedance piping. The remaining parameters are conventional parameters. Since the weak fault sensitive parameters are highly sensitive to weak faults, the sampling frequency of the weak fault sensitive parameters is increased. The parameter sampling frequency adjustment process is as follows: Set the standard value of the sampling frequency, set the sampling frequency of the weak fault sensitive parameters to 1.2 times the standard value, and set the sampling frequency of the regular parameters to 0.8 times the standard value. The collected data is subjected to adaptive Kalman filtering, moving average filtering, and 3σ outlier removal to suppress noise and interference, and finally timestamp alignment is performed.

[0019] S2. Screen the soft sensing estimation parameters for early fault warning and use the screened soft sensing estimation parameters as the relevant early warning parameters for the fault type. The soft sensing estimation parameters can be calculated from the collected multi-dimensional parameters and can be divided into lithium ion soft sensing estimation parameters, sodium ion soft sensing estimation parameters and vanadium liquid soft sensing estimation parameters. The parameters estimated by lithium-ion soft sensing include: cell internal temperature, cell state of charge, cell health status, and equivalent polarization voltage; the equivalent polarization voltage is the voltage component of the off-circuit voltage caused by electrochemical polarization and concentration polarization. The parameters estimated by sodium ion soft sensing include: cell internal temperature, cell state of charge, cell health status, dynamic equivalent internal resistance, sodium ion migration impedance, and gas evolution characteristic coefficient. Sodium ion migration impedance refers to the impedance of sodium ions during migration and deintercalation at the electrode or electrolyte interface, reflecting the electrode dynamics performance. The gas evolution characteristic coefficient is a characteristic coefficient that characterizes the gas generation rate and degree during the charging, discharging, and aging processes of the cell, and is used for safety early warning. The parameters estimated by the soft sensing of the vanadium-containing liquid include: equivalent internal resistance of the stack, state of charge of the electrolyte, membrane leakage coefficient, vanadium ion concentration difference between the positive and negative electrodes, average polarization voltage of the stack, and equivalent flow resistance of the pipeline. The average polarization voltage of the stack refers to the total polarization voltage caused by electrochemical polarization and concentration polarization of the stack under the operating current, which is used for efficiency analysis and control. Early failures can be categorized as: lithium-ion failures, sodium-ion failures, and all-vanadium liquid flow failures. Lithium-ion faults include: micro-short circuits, localized overheating, abnormally increased polarization, and slow capacity drift. Sodium ion failures include: increased interfacial impedance, impaired sodium ion migration, early gas evolution, abnormal local temperature rise, deterioration of voltage consistency, and accelerated cycle decay. All-vanadium redox flow faults include: membrane series leakage, increased stack internal resistance, electrolyte imbalance, decreased pipeline efficiency, abnormal polarization, uneven flow, and local blockage. The selection criteria are linear correlation values, nonlinear correlation values, and fault sensitivity values. For any type of fault, the process for selecting the relevant soft-sensor estimation parameters is as follows: The linear correlation value is used to measure the strength of the linear correlation between soft sensing estimation parameters and the degree of fault. The expression for the linear correlation value is as follows: (1); in, The linear correlation value, For covariance, Standard deviation, For the first Time series of soft sensing estimation parameters For the first Time series of severity scores for each fault type, and The timing is corresponding; in the preferred scheme, a severity score value rule is given, specifically: normal state, severity score is 0; minor early failure, severity score is 0.2; obvious early failure, severity score is 0.5; about to deteriorate, severity score is 0.8; obvious failure, severity score is 1. Nonlinear correlation values ​​are used to capture early nonlinear, coupled associations between soft-sensor estimation parameters and faults. The expression for nonlinear correlation values ​​is as follows: (2); in, This is a non-linear correlation value. The value of the parameters estimated by the soft sensing at a certain moment in the time series. The value at a specific moment in the time series representing the severity score of the fault type. for The ratio of the number of occurrences to the total number of elements in the time series. for The ratio of the number of occurrences to the total number of elements in the time series. For the same time, the soft sensing estimation parameters are taken Severity score The ratio of the number of occurrences to the total number of elements in the time series; the summation in equation (2), the number of elements to be summed is and The total number of combinations, which is the total number of elements minus the number of repetitions; for example... When taking 2, If 0.5 appears twice, that is, it appears once repeatedly, so the number of combinations should be reduced by 1. The fault sensitivity value is used to ensure that the parameter is sensitive to early faults but not sensitive to normal aging or fluctuations in operating conditions. The expression for the fault sensitivity value is as follows: (3); in, For the first The first soft sensing estimation parameter is related to the first... Fault sensitivity values ​​for different fault types Estimate the change in parameters for the corresponding soft sensing, i.e. The difference between the maximum and minimum values. This represents the change under normal aging conditions. This refers to the change under fluctuating operating conditions; The screening process is as follows: Set linear judgment threshold, nonlinear judgment threshold and sensitivity judgment threshold, and use the soft sensing estimation parameters that simultaneously satisfy: linear correlation value greater than linear judgment threshold, nonlinear correlation value greater than nonlinear judgment threshold, and fault sensitivity value greater than sensitivity judgment threshold as the relevant early warning parameters for this fault type.

[0020] S3. Perform early fault diagnosis. The detailed process is as follows: S301. For any fault type of any battery, set a warning threshold for each relevant warning parameter; the warning threshold is a dynamic threshold, and the expression is as follows: (4); in, As the warning threshold, For early warning adjustment items, As the baseline value for early warning, , These are the adjustment weight and the base weight, respectively, and their sum is 1. The expression for the early warning adjustment item is as follows: (5); in, This refers to the fault sensitivity value of the relevant warning parameter for this type of fault. This is the amplitude adjustment coefficient, and its value can be different for different relevant early warning parameters; The baseline value for early warning is determined based on the values ​​of relevant early warning parameters during fault-free operation. Specifically, it involves retrieving several sets of time-series data, each set containing several elements, using the current time as the reference. The average of the maximum values ​​of the elements in each set is then calculated, and this value is the baseline value for early warning. S302. Perform early fault judgment; the condition for early fault judgment is: when at least two relevant warning parameters exceed the warning threshold, it is judged as an early fault.

[0021] S4. Perform fault location and classification operations; early fault location uses a classification method, which can be divided into: cluster-level location, module-level location, and cell-level location; cluster-level location is judged based on cluster-level anomaly score, and the expression for the cluster-level anomaly score is as follows: (6); in, For cluster-level anomaly scores, For the first The average value of relevant early warning parameters over several sampling periods; for the function , there is: when When the warning threshold of the relevant warning parameter is not exceeded, Take 0, when When the warning threshold of the relevant warning parameter is exceeded Pick The absolute value of the difference between the warning threshold and the relevant warning parameter; Compare the cluster-level anomaly scores of all clusters, and the cluster with the highest cluster-level anomaly score is the cluster where the fault occurs; After locating the cluster where the fault occurs, module-level localization is performed. Module-level localization is determined based on the module anomaly score, which is expressed as follows: (7); in, For module anomaly scores, , These are the standard deviations of cell voltage and temperature within the module, respectively. The relative rate of change of the module's equivalent internal resistance is calculated by comparing the difference between the module's current internal resistance and its initial internal resistance with the initial internal resistance. , , These are the voltage evaluation weight, temperature evaluation weight, and internal resistance evaluation weight of the module, respectively. Compare the module anomaly scores of all modules, and the cluster with the highest module anomaly score is the module where the fault occurs; After locating the faulty module, cell-level localization is performed. Cell-level localization is based on cell anomaly scores, which are expressed as follows: (8); in, Score for cell abnormalities. , , These are the cell's voltage deviation rate, temperature rise rate, and relative change rate of internal resistance, respectively. , , These are the cell's voltage weight, temperature weight, and internal resistance weight, respectively. The temperature rise rate needs to be normalized, and its value range needs to be adjusted to 0~1. The voltage deviation rate of the cell can be obtained by comparing the difference between the cell voltage and the module's average voltage with the module's average voltage. Comparing the cell anomaly scores of all cells, the cell with the highest anomaly score is the cell that failed early.

[0022] S5. Implement early fault warning; an early fault warning uses a tiered warning method, specifically: set an anomaly threshold. When the cell anomaly score of an early faulty cell exceeds the threshold, a message is pushed to the power plant operation and maintenance manager to handle the early faulty cell; when the cell anomaly score of an early faulty cell does not exceed the threshold, a message is pushed to the power plant operation and maintenance manager, and the sampling frequency of the cell data is increased. An observation period is set. If the cell anomaly score of the cell does not improve during the observation period, the normal sampling frequency is restored; otherwise, a message is pushed to the power plant operation and maintenance manager to handle the cell.

[0023] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A novel early warning method for large-scale electrochemical energy storage faults, characterized in that, Includes the following steps: S1. Collect multidimensional relevant parameters and perform preprocessing operations; S2. Filter the soft sensor estimation parameters for early fault warning and use the filtered soft sensor estimation parameters as relevant early warning parameters for fault types. S3. Perform early fault diagnosis; S4. Perform early fault location and classification operations; S5. Provide early fault warnings.

2. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 1, characterized in that, The data collection targets three types of energy storage units: lithium-ion battery systems, sodium-ion battery systems, and vanadium redox flow battery systems. The parameters collected by lithium-ion battery systems and sodium-ion battery systems can be categorized into: electrical parameters, thermal parameters, gas parameters, and mechanical parameters. The parameters collected by the vanadium redox flow battery system can be divided into: stack electrical parameters, fluid parameters, ion concentration parameters, and equipment status parameters. The parameters are divided into sensitive parameters for minor faults and normal parameters.

3. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 2, characterized in that, The parameter sampling frequency adjustment process is as follows: Set the standard value of the sampling frequency, set the sampling frequency of the weak fault sensitive parameter to 1.2 times the standard value, and set the sampling frequency of the normal parameter to 0.8 times the standard value.

4. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 1, characterized in that, Soft sensing estimation parameters are divided into lithium-ion soft sensing estimation parameters, sodium-ion soft sensing estimation parameters, and all-vanadium liquid soft sensing estimation parameters. Early failures can be categorized as: lithium-ion failures, sodium-ion failures, and all-vanadium liquid flow failures.

5. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 4, characterized in that, The selection criteria are linear correlation values, nonlinear correlation values, and fault sensitivity values. For any type of fault, the process for selecting the relevant soft-sensor estimation parameters is as follows: The expression for the linear correlation value is as follows: （1）； in, The linear correlation value, For covariance, Standard deviation For the first Time series of soft sensing estimation parameters For the first Time series of severity scores for each fault type, and They correspond in time sequence; The expression for the nonlinear correlation value is as follows: （2）； in, This is a non-linear correlation value. The value of the parameters estimated by the soft sensing at a certain moment in the time series. The value at a specific moment in the time series representing the severity score of the fault type. for The ratio of the number of occurrences to the total number of elements in the time series. for The ratio of the number of occurrences to the total number of elements in the time series. For the same time, the soft sensing estimation parameters are taken Severity score The ratio of the number of occurrences to the total number of elements in the time series; The expression for the fault sensitivity value is as follows: （3）； in, For the first The first soft sensing estimation parameter is related to the first... Fault sensitivity values ​​for different fault types Estimate the change in parameters for the corresponding soft sensing, i.e. The difference between the maximum and minimum values. This represents the change under normal aging conditions. This represents the change under fluctuating operating conditions.

6. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 5, characterized in that, The screening process is as follows: Set linear judgment threshold, nonlinear judgment threshold and sensitivity judgment threshold, and use the soft sensing estimation parameters that simultaneously satisfy: linear correlation value greater than linear judgment threshold, nonlinear correlation value greater than nonlinear judgment threshold, and fault sensitivity value greater than sensitivity judgment threshold as the relevant early warning parameters for this fault type.

7. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 5, characterized in that, The detailed process of step S3 is as follows: S301. For any fault type of any battery, set a warning threshold for each relevant warning parameter. S302. Perform early fault judgment; the condition for early fault judgment is: when at least two relevant warning parameters exceed the warning threshold, it is judged as an early fault.

8. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 7, characterized in that, The warning threshold is a dynamic threshold, expressed as follows: （4）； in, As the warning threshold, For early warning adjustment items, As the baseline value for early warning, , These are the adjustment weights and the base weights, respectively. The expression for the early warning adjustment item is as follows: （5）； in, This refers to the fault sensitivity value of the relevant early warning parameter for this type of fault. This is the amplitude adjustment coefficient; The baseline value for the early warning is determined based on the values ​​of the relevant early warning parameters during fault-free operation.

9. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 8, characterized in that, Early fault location uses a hierarchical location method, which can be divided into: cluster-level location, module-level location, and cell-level location. Cluster-level localization is determined based on cluster-level anomaly scores, which are expressed as follows: （6）； in, For cluster-level anomaly scores, For the first The average value of relevant early warning parameters over several sampling periods; for the function , there is: when When the warning threshold of the relevant warning parameter is not exceeded, Take 0, when When the warning threshold of the relevant warning parameter is exceeded Pick The absolute value of the difference between the warning threshold and the relevant warning parameter; Compare the cluster-level anomaly scores of all clusters, and the cluster with the highest cluster-level anomaly score is the cluster where the fault occurs; After locating the cluster where the fault occurs, module-level localization is performed. Module-level localization is determined based on the module anomaly score, which is expressed as follows: （7）； in, For module anomaly scores, , These are the standard deviations of cell voltage and temperature within the module, respectively. This represents the relative rate of change of the module's equivalent internal resistance. , , These are the voltage evaluation weight, temperature evaluation weight, and internal resistance evaluation weight of the module, respectively. Compare the module anomaly scores of all modules, and the cluster with the highest module anomaly score is the module where the fault occurs; After locating the faulty module, cell-level localization is performed. Cell-level localization is based on cell anomaly scores, which are expressed as follows: （8）； in, Score for cell abnormalities. , , These are the cell's voltage deviation rate, temperature rise rate, and relative change rate of internal resistance, respectively. , , These are the voltage weight, temperature weight, and internal resistance weight of the battery cell, respectively. Comparing the cell anomaly scores of all cells, the cell with the highest anomaly score is the cell with an early failure.

10. The method for early warning of faults in large-scale novel electrochemical energy storage according to claim 9, characterized in that, Early fault warning adopts a tiered warning method, specifically: an anomaly threshold is set. When the cell anomaly score of an early faulty cell exceeds the threshold, a message is pushed to the power plant operation and maintenance manager to carry out operation and maintenance processing on the early faulty cell; when the cell anomaly score of an early faulty cell does not exceed the threshold, a message is pushed to the power plant operation and maintenance manager, and the sampling frequency of the cell data is increased. An observation period is set. If the cell anomaly score of the cell does not improve during the observation period, the normal sampling frequency is restored; otherwise, a message is pushed to the power plant operation and maintenance manager to carry out operation and maintenance processing on the cell.

Images